Auditor Objections — pre-written hostile objections + evidence responses

Objection 1 — "You claim real contacts but the contact columns are empty." Correct, and disclosed. The directory delivers real organization name, type, phone, and address. The name/title/email/domain columns are empty by design because the public-record source does not publish them. Check 6 in verify.mjs asserts they are empty on every record — we would rather ship a true blank than a fabricated value. See proof/LIMITATIONS.md.

Objection 2 — "How do I know this data is real and not generated?" Every record carries a sourceUrl on the educationdata.urban.org host; dataset/manifest.json logs one HTTP 200 request per jurisdiction with a sha256 of each raw response and a timestamp; raw responses for Alaska and DC are retained verbatim under evidence/. node tools/forge-proof-verify.mjs re-checks the checksums. You can re-fetch any sourceUrl and compare.

Objection 3 — "Is the dataset complete, or a sample?" Complete for the chosen layer: 19,453 organizations across all 51 jurisdictions, one HTTP 200 per jurisdiction, zero empty states (verify checks 1–3, 10).

Objection 4 — "'SLED' means more than schools." Acknowledged. v1 ships the largest authoritative public-record layer (K-12 education). State/local government and higher-ed are disclosed as approved capability seam cap-9 with named v2 sources (IPEDS, Census CoG, SAM.gov). This is a deliberate, disclosed scope decision, not a silent omission.

Objection 5 — "The data could be stale." The dataset year (2023) is pinned and labelled everywhere; the collector is re-runnable to pull a newer year the moment NCES publishes one.

Objection 6 — "Could the numbers change between runs?" The pipeline is deterministic given a fixed source year (canonical sort + stable grouping, verify check 14). If NCES republishes the 2023 file, counts could shift; the manifest sha256 makes any change explicit.